Skip to content
Docs
HomepageDemo
Overview
Getting started
QuickstartChoose an installationWhat is Metis?Accounts and initial setupConnect your first providerYour first taskSign inInstall on LinuxInstall on macOSInstall on Windows
Chat & context
Navigation & chatsCompose messagesResponses & tool callsAnswer questionsApprove actionsSearch & quick actions
Workspace
Agent overviewCanvasPlansMemoryFilesTerminalBrowserDesktopsServer monitorArrange the workspaceExport workspaces
Projects & teams
Create a projectProject overviewChat visibilityProject picture, color, and iconProject instructionsProject memoryProject skillsProject filesProject notesAgent model defaultLet agents manage the teamMembers & invitationsAgent teamsTeam presetsTeam overview
Notes
Notes & scratchpad
Agent Hub
SkillsConnections / MCPsCustom modes & rulesGlobal agent rules
Settings
GeneralModels — ConnectionsAgent — Runtime & memoryDevices — Remote clientsAdmin / ChatsUpdatesSearch settingsAttachments and file sizeSubagent modelToken compressionPreview compressionSounds and notificationsNotification deliverySet up voice inputChoose a microphoneRecording and streamingDictionaryConfigure the browser previewManage website dataWebsite and source code
Models & providers
AgentDeep ResearchPlanAskGauntletCursorOpenAIAnthropicGoogle GeminixAI / GrokOpenRouterOllama / LocalOpenAI-compatibleOpenAI CodexClaude CodeGoogle AntigravityGrok BuildOpenCodeModel options
Account & sharing
ProfileNotificationsUsage & costsShare chatsPublic shares
Guides & operations
Charts, graphs & mediaAutomationsOperate your installationTroubleshootingSet up a complete projectWork on a connected PCResearch and produce a reportGlossary and scopes
Metis AI
Docs/Settings

Settings

Token compression

Compression shortens redundant tool output and context before they reach the model.

1 min readUpdated: October 2026
General settings in the test instance. Captured in an isolated Metis test instance with fictional sample data (October 2026).
General settings in the test instance

General settings in the test instance. Captured in an isolated Metis test instance with fictional sample data (October 2026).

Step by step

  1. 01

    Enable Token compression.

  2. 02

    Choose Lite, Standard, Aggressive, RTK, Stacked, or Ultra.

  3. 03

    Use Tool results and Chat history to select which content is processed.

  4. 04

    Test the effect with Analyze preview using a representative sample.

Token compression

  • Enable
  • Lite/Standard/Aggressive/RTK/Stacked/Ultra
  • Tool results
  • Chat history

Good to know

  • Stronger levels can reduce wording and detail.
  • Fewer tokens do not automatically improve responses. Choose a level suited to your material.

Troubleshooting

If a setting has no effect, check its scope. Running tasks or existing uploads may retain their previous state.

Troubleshooting guide

Related articles

GeneralModels — ConnectionsAgent — Runtime & memoryDevices — Remote clients
PreviousSubagent modelNextPreview compression
Back to homepageReport an issue

On this page

Step by stepToken compressionGood to knowTroubleshooting
Back to top ↑
Documentation
Overview
Getting started
QuickstartChoose an installationWhat is Metis?Accounts and initial setupConnect your first providerYour first taskSign inInstall on LinuxInstall on macOSInstall on Windows
Chat & context
Navigation & chatsCompose messagesResponses & tool callsAnswer questionsApprove actionsSearch & quick actions
Workspace
Agent overviewCanvasPlansMemoryFilesTerminalBrowserDesktopsServer monitorArrange the workspaceExport workspaces
Projects & teams
Create a projectProject overviewChat visibilityProject picture, color, and iconProject instructionsProject memoryProject skillsProject filesProject notesAgent model defaultLet agents manage the teamMembers & invitationsAgent teamsTeam presetsTeam overview
Notes
Notes & scratchpad
Agent Hub
SkillsConnections / MCPsCustom modes & rulesGlobal agent rules
Settings
GeneralModels — ConnectionsAgent — Runtime & memoryDevices — Remote clientsAdmin / ChatsUpdatesSearch settingsAttachments and file sizeSubagent modelToken compressionPreview compressionSounds and notificationsNotification deliverySet up voice inputChoose a microphoneRecording and streamingDictionaryConfigure the browser previewManage website dataWebsite and source code
Models & providers
AgentDeep ResearchPlanAskGauntletCursorOpenAIAnthropicGoogle GeminixAI / GrokOpenRouterOllama / LocalOpenAI-compatibleOpenAI CodexClaude CodeGoogle AntigravityGrok BuildOpenCodeModel options
Account & sharing
ProfileNotificationsUsage & costsShare chatsPublic shares
Guides & operations
Charts, graphs & mediaAutomationsOperate your installationTroubleshootingSet up a complete projectWork on a connected PCResearch and produce a reportGlossary and scopes

Popular starting points

Getting startedQuickstartGo from installation to your first verifiable task: set up Metis, connect a model, and start with a small request.Getting startedChoose an installationNative installation is recommended for Linux, macOS, and Windows. Install Metis on the computer or server that will run it, then pGetting startedWhat is Metis?Metis is your self-hosted workspace for AI agents. Conversations, files, browser, tools, and persistent context stay together. YouGetting startedAccounts and initial setupOn first launch, you set up the instance with an administrator account. The single-use operator token confirms your authority to mGetting startedConnect your first providerMetis needs model access to generate responses and run agent tasks. Setup displays authentication methods supported by the selecteGetting startedYour first taskCheck your setup with a small task whose result you can assess directly. You can then add files, browser access, or project contexGetting startedSign inSign in with an account on the relevant Metis instance. Your displayed profile name may differ from your login username.
↑ ↓ select · Enter openEsc close